LLM-backed chat via HF Inference (Gemma 3 27B) with fixed rule-based fallback

Fix PDF report cursor bug; add robust file loading and richer preprocessing options

Upgrade to Gradio 5.50 (4.44 incompatible with current dependency stack)

Disable ChatInterface example caching (broken with messages-format chatbot on 4.44)

Initial deploy: ML Data Analysis Studio